Summary
The episode covers Perplexity's launch of Perplexity Max, a cloud-based agent that orchestrates 19 different AI models to execute multi-step workflows and can control a user's screen for automation. The host discusses a pulled live demo that highlighted how brittle and fast-moving advanced agent demos remain. Perplexity's strategic shift away from ad-driven scale toward a narrower professional audience (targeting 'GDP-moving' decisions) and premium pricing is examined. The company's multimodal, multi-LLM approach — routing queries to specialized models based on cost and performance and spawning sub-agents for sub-problems — is explained alongside concerns about model sourcing and disclosure.
Key Takeaways
- 1Perplexity Max is a cloud-hosted multi-agent system that coordinates 19 AI models to automate complex, multi-step workflows.
- 2Perplexity prioritizes a smaller, professional user base over ad-driven scale, shifting to premium pricing and enterprise-focused functionality.
- 3The system dynamically routes queries across specialized LLMs using cost and performance tradeoffs and runs internal benchmarks to guide selection.
- 4Agent demos remain fragile; Perplexity pulled a live demo due to last-minute software flaws, illustrating the risks of public agent demonstrations.
- 5Using modified open-source and international LLMs can lower costs but raises disclosure and data-routing concerns.
Notable Quotes
"They said that their premium subscribers are gonna have access to a new agent tool that can take 19 different AI models all working together in computer use meaning it can kind of take control of your screen and get jobs done for you."
"They quote, unifies every current AI capability into a single system."
"They've reported that they have 900 million weekly chatGPT users that are all going to, I mean, a lot of the free ones are going to start getting seen kind of these ads everywhere."
"Their system right now can automatically choose what model based on cost and performance."
Episode questions
What is Perplexity Max and how does it differ from local device agents like OpenRouter/OpenClaw?
Perplexity Max is a cloud-based agent that orchestrates 19 AI models and can execute complex workflows, including controlling the screen; because it's cloud-hosted Perplexity claims it avoids many security risks associated with local device agents. The cloud architecture also enables spawning sub-agents and centralized orchestration.
Who is Perplexity targeting with its new strategy and pricing?
Perplexity is targeting a narrower professional audience (enterprise and 'GDP-moving' decision-makers) rather than mass-market users; their highest tier (Perplexity Max) is premium-priced (~$200/month) and oriented to professionals needing deep research and workflow automation.
How does Perplexity decide which model to use for a given task?
Their system routes queries across multiple LLMs and selects models based on cost and performance tradeoffs, using specialized models for domains (e.g., Gemini for vision, Claude for engineering, GPT for medical research). They also run internal benchmarks (Draco) to compare complex research capabilities.
Did Perplexity publicly demo Perplexity Max and what happened?
They planned a live demo but pulled it after discovering last-minute software flaws, illustrating the fragility and rapid pace of development for agent products; examples of other companies (e.g., Google Glass demos) were cited to contextualize such failures.